SpatialTree: How Spatial Abilities Branch Out in MLLMs

TL;DR

SpatialTree introduces a hierarchical framework inspired by cognitive science to evaluate and enhance spatial abilities in multimodal large language models, revealing interdependencies and transfer dynamics across different ability levels.

Contribution

It presents the first capability-centric hierarchy and benchmark for spatial abilities in MLLMs, along with insights into transfer dynamics and a novel RL strategy for improvement.

Findings

L1 skills are largely orthogonal, higher skills are correlated.

Negative transfer occurs within L1, but positive transfer from low to high levels.

Auto-think strategy improves performance across all hierarchy levels.

Abstract

Cognitive science suggests that spatial ability develops progressively-from perception to reasoning and interaction. Yet in multimodal LLMs (MLLMs), this hierarchy remains poorly understood, as most studies focus on a narrow set of tasks. We introduce SpatialTree, a cognitive-science-inspired hierarchy that organizes spatial abilities into four levels: low-level perception (L1), mental mapping (L2), simulation (L3), and agentic competence (L4). Based on this taxonomy, we construct the first capability-centric hierarchical benchmark, thoroughly evaluating mainstream MLLMs across 27 sub-abilities. The evaluation results reveal a clear structure: L1 skills are largely orthogonal, whereas higher-level skills are strongly correlated, indicating increasing interdependency. Through targeted supervised fine-tuning, we uncover a surprising transfer dynamic-negative transfer within L1, but strong cross-level transfer from low- to high-level abilities with notable synergy. Finally, we explore how to improve the entire hierarchy. We find that naive RL that encourages extensive "thinking" is unreliable: it helps complex reasoning but hurts intuitive perception. We propose a simple auto-think strategy that suppresses unnecessary deliberation, enabling RL to consistently improve performance across all levels. By building SpatialTree, we provide a proof-of-concept framework for understanding and systematically scaling spatial abilities in MLLMs.